The python API for Micron DLA has these functions:
Compiles a ONNX/NNEF network and produce .bin file with everything that is needed to execute
Parameters:
modelpath: path to a model file in ONNX format.
outfile: path to a file where a model in Micron DLA ready format will be saved. If this param is used then Init call is needed afterwards
inshapes: it is an optional string with shape information in the form of size0xsize1x...sizeN. In case of multiple inputs, shapes are semi-colon separated. This parameter is normally inferred from the model file, it can be overridden in case we want to change some input dimension
samples: a list of images in numpy float32 format used to choose the proper quantization for variable-fixed-point
MDLA: MDLA object to link together so that models can be load into memory together
Return value: List of the output nodes names returned by the network
Loads a bitfile on an FPGA if necessary and prepares to run Micron DLA hardware.
Parameters:
infile: model binary file path. .bin file created by Compile
MDLA: another MDLA obj to be combined with this MDLA run.
Set some flags that change the behaviour of the API.
Parameters:
Name name of the flag to be set
Value value to set the flag as a numpy string
Currently available options are listed in here
Gets information of the SDK options.
Parameters:
Name info name to be returned
Currently available options are listed in here
Frees the network.
Parameters:
None
Get an output from a buffer. If the blocking flag was set then it will wait for Micron DLA hardware.
Return value:: Output tensor or list of output tensors and the userobj that was associated with this
buffer in the PutInput function call.
Put an input into a buffer and start Micron DLA hardware.
Parameters:
Image input data as a [list of] numpy array of type float32
userobj user defined object to keep track of the given input
Return value: Error or no error
Runs a single inference on Micron DLA hardware.
Parameters:
Image input data as a [list of] numpy array of type float32
Return Result output tensor or list of output tensors of the model
Runs a single inference on the Micron DLA hardware simulator.
Parameters:
Image input data as a [list of] numpy array of type float32
Return Result output tensor or list of output tensors of the model
Runs a single inference using thnets.
Parameters:
Image input data as a [list of] numpy array of type float32
Return Result output tensor or list of output tensors of the model